Skip to content

Short-Term Reversal Strategy

Profit from weekly price reversals.

Strategy Overview

Stocks that performed poorly last week tend to outperform next week, and vice versa. This is one of the most robust anomalies in finance.

The Signal

Text Only
signal short_term_reversal:
  // Last week's return
  ret_5d = ret(prices, 5)

  // Negative return = buy (expect reversal up)
  emit -zscore(ret_5d)

Complete Strategy

Text Only
data:
  source = "prices_with_sectors.parquet"
  format = parquet

// Short-term reversal
signal str:
  // 5-day return
  ret_5d = ret(prices, 5)

  // Revert (buy losers, short winners)
  emit neutralize(-zscore(ret_5d), by=sectors)

portfolio short_term_reversal:
  weights = rank(str).long_short(
    top = 0.2,
    bottom = 0.2,
    cap = 0.03
  )

  constraints:
    gross_exposure = 2.0
    net_exposure = 0.0
    max_sector = 0.20

  costs = tc.bps(10)

  // Weekly rebalancing
  backtest rebal=5 from 2015-01-01 to 2024-12-31

Why It Works

  1. Liquidity shocks: Forced sellers push prices down temporarily
  2. Overreaction: Investors overreact to short-term news
  3. Bid-ask bounce: Prices bounce between bid and ask
  4. Market microstructure: Short-term noise

Variations

Volume-Weighted Reversal

Text Only
signal volume_weighted:
  ret_5d = ret(prices, 5)

  // High volume = more conviction in reversal
  vol_ratio = volume / rolling_mean(volume, 20)
  high_volume = vol_ratio > 1.5

  // Stronger signal when volume is high
  signal = -ret_5d * where(high_volume, 1.5, 1.0)

  emit zscore(signal)

Idiosyncratic Reversal

Text Only
signal idio_reversal:
  // Remove market component
  stock_ret = ret(prices, 5)
  market_ret = ret(market, 5)

  // Idiosyncratic return
  idio_ret = stock_ret - market_ret

  // Revert idiosyncratic component
  emit -zscore(idio_ret)

Multi-Day Reversal

Text Only
signal multi_day:
  // Multiple short-term windows
  ret_1d = ret(prices, 1)
  ret_3d = ret(prices, 3)
  ret_5d = ret(prices, 5)

  // Weighted combination
  combined = -0.2 * zscore(ret_1d) - 0.3 * zscore(ret_3d) - 0.5 * zscore(ret_5d)

  emit neutralize(combined, by=sectors)

Extreme Moves Only

Text Only
signal extreme_reversal:
  ret_5d = ret(prices, 5)

  // Only trade extreme moves
  extreme_down = ret_5d < quantile(ret_5d, 0.1)  // Bottom 10%
  extreme_up = ret_5d > quantile(ret_5d, 0.9)    // Top 10%

  signal = where(extreme_down, -ret_5d,
           where(extreme_up, -ret_5d, 0))

  emit zscore(signal)

Liquidity Filter

Text Only
signal liquid_reversal:
  ret_5d = ret(prices, 5)

  // Only trade liquid stocks
  adv = rolling_mean(volume * prices, 20)
  liquid = adv > 1000000  // $1M+ daily volume

  signal = where(liquid, -ret_5d, 0)

  emit zscore(signal)

Intraday vs Overnight

Text Only
signal overnight_reversal:
  // Overnight return (close to open)
  overnight_ret = (open - lag(close, 1)) / lag(close, 1)

  // Intraday return (open to close)
  intraday_ret = (close - open) / open

  // Overnight moves revert more
  emit -zscore(overnight_ret)

With Momentum Filter

Avoid reverting stocks with strong momentum:

Text Only
signal filtered_reversal:
  // Short-term reversal
  ret_5d = ret(prices, 5)

  // Long-term momentum
  ret_60d = ret(prices, 60)

  // Don't fight strong momentum
  strong_up = ret_60d > quantile(ret_60d, 0.8)
  strong_down = ret_60d < quantile(ret_60d, 0.2)

  // Reduce reversal in momentum stocks
  signal = -ret_5d
  signal = where(strong_up and ret_5d < 0, signal * 0.5, signal)
  signal = where(strong_down and ret_5d > 0, signal * 0.5, signal)

  emit zscore(signal)

Expected Results

Text Only
Backtest Results: short_term_reversal
=====================================
Period: 2015-01-01 to 2024-12-31

Returns:
  Total Return: 38%
  Annual Return: 3.4%
  Annual Volatility: 6.8%
  Sharpe Ratio: 0.50

Turnover:
  Annual Turnover: 1040%
  Avg Holding Period: 5 days

Characteristics:
  Win Rate: 52%
  Profit Factor: 1.15

Transaction Costs

High turnover means costs matter significantly:

Text Only
Gross Return: 5.2%
Costs (10bps): -1.0%
Costs (20bps): -2.1%
Net Return: 3.1% - 4.2%

Risk Considerations

Momentum Stocks

Reversal fails in momentum stocks:

Text Only
// Exclude strong momentum
signal safe_reversal:
  str = -ret(prices, 5)
  mom = ret(prices, 60)

  // Exclude top/bottom momentum quintiles
  strong_mom = abs(mom) > quantile(abs(mom), 0.8)

  emit where(strong_mom, 0, zscore(str))

Event Risk

Earnings, M&A can cause permanent moves:

Text Only
// Exclude around earnings
signal event_filtered:
  str = -ret(prices, 5)

  // Avoid if big move (might be event)
  big_move = abs(ret(prices, 1)) > 0.10

  emit where(big_move, 0, zscore(str))

See Also